AB-Training: A Communication-Efficient Approach for Distributed Low-Rank Learning

TL;DR

AB-training is a communication-efficient distributed training method that uses low-rank representations and independent groups to significantly reduce network traffic, improve scalability, and enhance generalization in neural network training.

Contribution

This paper introduces AB-training, a novel low-rank, data-parallel approach that reduces communication overhead and improves training efficiency in distributed neural network training.

Findings

Reduced network traffic by approximately 70.31% across various scenarios

Achieved a 44.14:1 compression ratio on VGG16 with minimal accuracy loss

Outperformed traditional data parallel training by 1.55% on ResNet-50 with ImageNet

Abstract

Communication bottlenecks severely hinder the scalability of distributed neural network training, particularly in high-performance computing (HPC) environments. We introduce AB-training, a novel data-parallel method that leverages low-rank representations and independent training groups to significantly reduce communication overhead. Our experiments demonstrate an average reduction in network traffic of approximately 70.31\% across various scaling scenarios, increasing the training potential of communication-constrained systems and accelerating convergence at scale. AB-training also exhibits a pronounced regularization effect at smaller scales, leading to improved generalization while maintaining or even reducing training time. We achieve a remarkable 44.14 : 1 compression ratio on VGG16 trained on CIFAR-10 with minimal accuracy loss, and outperform traditional data parallel training by 1.55\% on ResNet-50 trained on ImageNet-2012. While AB-training is promising, our findings also reveal that large batch effects persist even in low-rank regimes, underscoring the need for further research into optimized update mechanisms for massively distributed training.